cleanup: optimize perfermance

[xu-cheng]

Only remove .DS_Store files when -s is passed.

On my Mac with 156 formulae installed, it will take about 10 seconds to
remove all the .DS_Store files

[xu-cheng]

I just realized that it's so slow on my Mac is because I have MacTeX installed. There're 143631 files/folders inside /usr/local/texlive.

[dplarson]

I've also got MacTeX installed and have noticed the same issue (i.e. that brew cleanup is slowed down in the removing .DS_Store files step). But perhaps it would be better to add a cleanup flag to ignore the .DS_Store removal step (instead of a flag to include removing .DS_Store) since most users probably don't have this issue.

[tdsmith]

Heh, same. Maybe skip the texlive path in rm_DS_Store instead?

[xu-cheng]

Heh, same. Maybe skip the texlive path in rm_DS_Store instead?

There could be variety non-homebrew paths inside HOMEBREW_PREFIX. So I think it's better to remove .DS_Store files only in selective paths.

[MikeMcQuaid]

Yeh, let's whitelist directories here instead of requiring a switch.

[xu-cheng]

Updated.

[MikeMcQuaid]

👍

[bfontaine]

You could give all paths to find to speed up the process, i.e. use only quiet_system call instead of one per directory.

Update: Something like:

paths = %w[Cellar Frameworks Library bin etc include lib opt sbin share var].map { |p| HOMEBREW_PREFIX/p }
paths.select!(&:exist?)

quiet_system "find", paths * " ", "-name", ".DS_Store", "-delete"

[xu-cheng]

You could give all paths to find to speed up the process

Why it will speed up? It's the same amount folders which will be scanned.

quiet_system "find", paths * " ", "-name", ".DS_Store", "-delete"

This wouldn't work. It should be:

args = paths
args << %w[-name .DS_Store -delete]
quiet_system "find", *args

[bfontaine]

Why it will speed up? It's the same amount folders which will be scanned.

Yes, but it fires find only once. Each call to quiet_system does the following:

1. fork the current process
2. redirect stdout and stderr to /dev/null
3. exec the command
4. wait for the command to end

find is waaay faster than Ruby since it’s compiled from optimized C. I just tested the following on my machine:

# create 1000 directories with a lot of subdirectories
$ for i in {1..1000}; do d=dir$i/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r; mkdir -p $d; touch $d/{x,y,z}; done

# search for all "x" files with `find`
$ time find dir{1..1000} -name x >/dev/null

real    0m0.272s
user    0m0.057s
sys 0m0.215s

# do the same in Ruby
$ brew irb
>>> b=Time.now;(1..1000).each { |i| quiet_system "find", "dir#{i}", "-name", "x" };Time.now-b
5.933559

That’s just a test, but it took 20+ times more time to do the loop in Ruby than letting find do it itself.

This wouldn't work.

Right, sorry.

[xu-cheng]

Updated